Re: [Qemu-devel] [PATCH 0/7] virtio: virtio-blk data plane

[Khoa Huynh]

Asias He <address@hidden> wrote on 11/20/2012 03:02:07 AM:

> From: Asias He <address@hidden>
> To: Stefan Hajnoczi <address@hidden>,
> Cc: address@hidden, Anthony Liguori/Austin/address@hidden, Paolo
> Bonzini <address@hidden>, Kevin Wolf <address@hidden>,
> "Michael S. Tsirkin" <address@hidden>, Khoa Huynh/Austin/address@hidden
> Date: 11/20/2012 03:01 AM
> Subject: Re: [PATCH 0/7] virtio: virtio-blk data plane
>
> Hello Stefan,
>
> On 11/15/2012 11:18 PM, Stefan Hajnoczi wrote:
> > This series adds the -device virtio-blk-pci,x-data-plane=on property that
> > enables a high performance I/O codepath.  A dedicated thread is
> used to process
> > virtio-blk requests outside the global mutex and without going
> through the QEMU
> > block layer.
> >
> > Khoa Huynh <address@hidden> reported an increase from 140,000
> IOPS to 600,000
> > IOPS for a single VM using virtio-blk-data-plane in July:
> >
> >   http://comments.gmane.org/gmane.comp.emulators.kvm.devel/94580
> >
> > The virtio-blk-data-plane approach was originally presented at
> Linux Plumbers
> > Conference 2010.  The following slides contain a brief overview:
> >
> >   http://linuxplumbersconf.org/2010/ocw/system/presentations/651/
> original/Optimizing_the_QEMU_Storage_Stack.pdf
> >
> > The basic approach is:
> > 1. Each virtio-blk device has a thread dedicated to handling ioeventfd
> >    signalling when the guest kicks the virtqueue.
> > 2. Requests are processed without going through the QEMU block layer using
> >    Linux AIO directly.
> > 3. Completion interrupts are injected via irqfd from the dedicated thread.
> >
> > To try it out:
> >
> >   qemu -drive if=none,id=drive0,cache=none,aio=native,format=raw,file=...
> >        -device virtio-blk-pci,drive=drive0,scsi=off,x-data-plane=on
>
>
> Is this the latest dataplane bits:
> (git://github.com/stefanha/qemu.git virtio-blk-data-plane)
>
> commit 7872075c24fa01c925d4f41faa9d04ce69bf5328
> Author: Stefan Hajnoczi <address@hidden>
> Date:   Wed Nov 14 15:45:38 2012 +0100
>
>     virtio-blk: add x-data-plane=on|off performance feature
>
>
> With this commit on a ramdisk based box, I am seeing about 10K IOPS with
> x-data-plane on and 90K IOPS with x-data-plane off.
>
> Any ideas?

Hi Asias, I'll try this latest patchset next week and will report results
as soon as they are available.  I am currently on vacation this week due
to the Thanksgivings holiday in the US.... (Previously, I was able to get
up to 1.33 million IOPS for a single KVM guest using Stefan's previous
code.)

Thanks,
-Khoa

>
> Command line I used:
>
> IMG=/dev/ram0
> x86_64-softmmu/qemu-system-x86_64 \
> -drive file=/root/img/sid.img,if=ide \
> -drive file=${IMG},if=none,cache=none,aio=native,id=disk1 -device
> virtio-blk-pci,x-data-plane=off,drive=disk1,scsi=off \
> -kernel $KERNEL -append "root=/dev/sdb1 console=tty0" \
> -L /tmp/qemu-dataplane/share/qemu/ -nographic -vnc :0 -enable-kvm -m
> 2048 -smp 4 -cpu qemu64,+x2apic -M pc
>
>
> >
> > Limitations:
> >  * Only format=raw is supported
> >  * Live migration is not supported
> >  * Block jobs, hot unplug, and other operations fail with -EBUSY
> >  * I/O throttling limits are ignored
> >  * Only Linux hosts are supported due to Linux AIO usage
> >
> > The code has reached a stage where I feel it is ready to merge.  Users have
> > been playing with it for some time and want the significant
> performance boost.
> >
> > We are refactoring QEMU to get rid of the global mutex.  I believe that
> > virtio-blk-data-plane can eventually become the default mode of operation.
> >
> > Instead of waiting for global mutex removal efforts to finish, I want to use
> > virtio-blk-data-plane as an example device for AioContext and threaded hw
> > dispatch refactoring.  This means:
> >
> > 1. When the block layer can bind to an AioContext and execute I/O
> outside the
> >    global mutex, virtio-blk-data-plane can use this (and gain image format
> >    support).
> >
> > 2. When hw dispatch no longer needs the global mutex we can use hw/virtio.c
> >    again and perhaps run a pool of iothreads instead of dedicated data plane
> >    threads.
> >
> > But in the meantime, I have cleaned up the virtio-blk-data-plane
> code so that
> > it can be merged as an experimental feature.
> >
> > Changes from the RFC v9:
> >  * Add x-data-plane=on|off option and coexist with regular virtio-blk code
> >  * Create thread from BH so it inherits iothread cpusets
> >  * Drain requests on vm_stop() so stopped guest does not access image file
> >  * Add migration blocker
> >  * Add bdrv_in_use() to prevent block jobs and other operations
> that can interfere
> >  * Drop IOQueue request merging for simplicity
> >  * Drop ioctl interrupt injection and always use irqfd for simplicity
> >  * Major cleanup to split up source files
> >  * Rebase from qemu-kvm.git onto qemu.git
> >  * Address Michael Tsirkin's review comments
> >
> > Stefan Hajnoczi (7):
> >   raw-posix: add raw_get_aio_fd() for virtio-blk-data-plane
> >   configure: add CONFIG_VIRTIO_BLK_DATA_PLANE
> >   dataplane: add virtqueue vring code
> >   dataplane: add event loop
> >   dataplane: add Linux AIO request queue
> >   dataplane: add virtio-blk data plane code
> >   virtio-blk: add x-data-plane=on|off performance feature
> >
> >  block.h                    |   9 +
> >  block/raw-posix.c          |  34 ++++
> >  configure                  |  21 +++
> >  hw/Makefile.objs           |   2 +-
> >  hw/dataplane/Makefile.objs |   3 +
> >  hw/dataplane/event-poll.c  | 109 ++++++++++++
> >  hw/dataplane/event-poll.h  |  40 +++++
> >  hw/dataplane/ioq.c         | 118 +++++++++++++
> >  hw/dataplane/ioq.h         |  57 +++++++
> >  hw/dataplane/virtio-blk.c  | 414 ++++++++++++++++++++++++++++++++
> +++++++++++++
> >  hw/dataplane/virtio-blk.h  |  41 +++++
> >  hw/dataplane/vring.c       | 321 +++++++++++++++++++++++++++++++++++
> >  hw/dataplane/vring.h       |  54 ++++++
> >  hw/virtio-blk.c            |  59 ++++++-
> >  hw/virtio-blk.h            |   1 +
> >  hw/virtio-pci.c            |   3 +
> >  trace-events               |   9 +
> >  17 files changed, 1293 insertions(+), 2 deletions(-)
> >  create mode 100644 hw/dataplane/Makefile.objs
> >  create mode 100644 hw/dataplane/event-poll.c
> >  create mode 100644 hw/dataplane/event-poll.h
> >  create mode 100644 hw/dataplane/ioq.c
> >  create mode 100644 hw/dataplane/ioq.h
> >  create mode 100644 hw/dataplane/virtio-blk.c
> >  create mode 100644 hw/dataplane/virtio-blk.h
> >  create mode 100644 hw/dataplane/vring.c
> >  create mode 100644 hw/dataplane/vring.h
> >
>
>
> --
> Asias
>